"Ty via Wine-Devel" <wine-devel@list.winehq.org> writes:
I wonder if the tests would be allowed into wine - wine has a strict cleanroom policy (https://gitlab.winehq.org/wine/wine/-/wikis/Clean-Room-Guidelines) which is to ensure its safety and longevity. Previously, Y. Osipov (mcm@etersoft.ru) had brought this up, though I did not see any replies on the mailing list on the subject.
In my mind, LLM's pull from all sorts of corners of the internet, without telling the user what was sourced from where. I would imagine (as NOT a lawyer) this has implications for licensing and plagiarism that we have yet to see play out in courts, as well as potentially running afoul the cleanroom policy (ie, see point 3 on the 'Donts' section about code that is available but not compatible, or possibly even leaked source). Given that a user has little to no control over source material and/or training data, I would personally assume this to mean it violates cleanroom by being unable to prove/control this - but I can also see some arguments going the other way depending on the apforforproach used.
I bring this up because I think we probably should figure out exactly what the policy on LLM use is/is going to be, and write it out clearly. What constitutes a cleanroom approach for one person may not be the same as another person's interpretation.
Clean room is not even the main issue. In order to ship any code, we need a proper authorship record, and an LGPL-compatible license. The legal copyright/licensing status of LLM output is currently not well defined, so we can't accept such code. -- Alexandre Julliard julliard@winehq.org